Models for BabyLM 2026: residual baselines, Block Attention Residuals, and static-routing controls trained on 10M and 100M words.
-
ypwhere/attnres4babylm2026-10m-l16-dynamic-s1337
Text Generation • 62.2M • Updated • 211 -
ypwhere/attnres4babylm2026-10m-l16-dynamic-s1338
Text Generation • 62.2M • Updated • 220 -
ypwhere/attnres4babylm2026-10m-l16-dynamic-s1339
Text Generation • 62.2M • Updated • 211 -
ypwhere/attnres4babylm2026-10m-l16-baseline-s1337
Text Generation • 62.2M • Updated • 220